Understanding the underlying reasons for potential human risky driving behaviors is crucial for improving road safety. Recent technologies allow the analysis of driving behaviors at a microscopic level, allowing a naturalistic observation of such phenomenon through information collected unobtrusively. This paper assesses the factors that influence discretionary lane changes on an urban highway in Santiago, Chile, employing an interpretable machine learning approach. We use full real-world vehicle-by-vehicle data gathered from Automatic Vehicle Identification technology and individual public information of the vehicles and their owners, which allows us to understand patterns that might have different characteristics when performed in simulated environments. We provide insights about the causes that increase the likelihood of lane changes. For example, we find that: (i) the older the car, the less likely it is to change lane, (ii) younger drivers change lane more often, and (iii) motorcycles drivers are the most likely to change lane.
- Automatic Vehicle Identification
- Drivers’ behaviors
- Interpretable Machine Learning
- lane change